Single Agent Robust Deep Reinforcement Learning for Bus Fleet Control
This research tackles bus bunching in complex transit systems using a single-agent deep reinforcement learning approach instead of traditional multi-agent methods. The core innovation is encoding vehicle and station IDs into the state space to allow one agent to manage the entire fleet effectively. A specially designed reward function balances schedule adherence and uniform headways, leading to a robust SAC model that eliminates bunching better than other advanced algorithms.